现代电子技术2025，Vol.48Issue(5)：43-48,6.DOI:10.16652/j.issn.1004-373x.2025.05.007
基于多分类自适应聚焦损失与B-CNN的棉田昆虫细粒度图像分类研究
Research on cotton insects' fine-grained image classification based on multi-class adaptive Focal Loss and B-CNN
摘要
Abstract
A research method based on multi-class adaptive Focal Loss function and bilinear convolutional neural network(B-CNN)is presented aiming at the fine-grained image classification of insects in cotton fields under complex background.B-CNN is selected as the backbone network,InceptionV3 is pre-trained as the feature extraction network,and CBAM(convolutional block attention module)module is added,so as to extract image features more effectively.A multi-class adaptive Focal Loss function is designed to improve the model's recognition ability to the few categories.In addition,L2 regularization is added to the process of model training to get rid of the overfitting,and the ReduceLROnPlateau learning rate scheduler is used to help the model reach the optimal solution.Experimental results show that the accuracy of the proposed model on the verification set reaches 97.52%,and its accuracy on the test set reaches 97.14% .Meanwhile,the evaluation indexes of both the loss value and the F1 score of the proposed model are better than those of the other comparison models.This study not only provides an effective technique for image classification of insects in cotton fields,but also provides a useful reference for fine-grained image classification in other fields.关键词
棉田昆虫/B-CNN/多分类自适应聚焦损失/InceptionV3/CBAM/细粒度图像分类Key words
cotton insect/B-CNN/multi-class adaptive Focal Loss/InceptionV3/CBAM/fine-grained image classification分类
电子信息工程引用本文复制引用
郝月华,吕卫东,张幽迪,冯俊磊..基于多分类自适应聚焦损失与B-CNN的棉田昆虫细粒度图像分类研究[J].现代电子技术,2025,48(5):43-48,6.基金项目
国家自然科学基金项目:多自由度非光滑脉冲耦合振子的锁频共振与参数振动:分析与计算(11962011) （11962011）
